Deployment and handover

Your best people, each with AI working for them, in three mornings.

I work with your team for four hours on a Tuesday, a Wednesday and a Thursday. By Thursday afternoon your people have AI clerks working for them on your own account, they can maintain and change it themselves, and the source code is yours to keep.

Request a week See it working

Where each step goes

Before anything is built, we map the work. Every step goes to one of three places, and every handoff gets a check. The tools will be different next year. Deciding where each step goes is the engineering, and it's what I bring.

A person

The judgment

The gray zone the rules can't settle. Anything that leaves the building, moves money past a limit, or needs a name on it. People approve, sign off and send.

An AI clerk

Three kinds of work

Reading text into fields, making a call with its confidence attached, and drafting words for a person to send. Nothing else.

The system

Everything code can do

Looking things up, adding up amounts, dates, matching, moving a record. If code can do a step, code does it — the same way every time, and never a guess.

And a check at every handoff. The system adds up what a clerk read and compares it with the total on the page. A call made with low confidence goes to a person. A draft carries the system's numbers, never the model's. When a check fails, the work stops and a person sees why.

The map moves. As models improve, a step can move from a person to a clerk — on evidence, and it moves back the moment its record slips. Your team leaves with the map and knows how to move it.

Watch a map move

The three mornings

We work together in the morning. Your team works on it alone for the rest of the day. The next morning starts with what they found.

Tuesday

Map it and stand it up

We map the work you chose: every step marked for a person, an AI clerk or the system, with the check at each handoff. Then we deploy it to your own account together, with your IT person at the keyboard. Your team signs in. We set up the first role — what it may touch, how far it may go, what waits for a person — and put the first AI clerk to work for one of your people.

That afternoon, on your ownYour team gives the clerk work in a practice company, reviews what it did, and writes down how the job is really done.

Wednesday

Make it yours

We start with Tuesday's questions. Then your people write down what they know about the job, and it becomes guidance their AI clerks follow. We test each piece against work already done before anyone relies on it, and whoever taught it gets the credit. We set the approval limits and the review routine for your real process.

That afternoon, on your ownYour IT person makes a change and deploys it without me. Your team teaches their clerks more of what they know.

Thursday

Hand it over

Questions first. Then the things you'll need after I've gone: adding a role, adding a tool, connecting another system, changing a model, backups, and reading the record when something looks wrong.

From then onIt's yours. Your people run it, and they've already changed it once without help.

Why mornings, and why three days. People learn a system by using it without the person who built it in the room. The afternoons are where that happens, and the next morning is where the real questions get answered.

It starts with your people, not your headcount

Most AI projects begin by asking which jobs can go. This one begins by asking who your best people are.

The usual pitchWhat you get here

"Which jobs can we cut?""Who does this best?" They teach it how the work is really done.

Success is hours saved.Success is work a person taught and a person approved — counted, by name.

The AI acts, and people find out afterwards.Every AI clerk works for a named person, who reviews its work until its record earns lighter review.

What it learns stays with the vendor.What your people know is written down in your own library, credited to whoever wrote it.

It runs on someone else's platform, on a monthly retainer.It runs on your own account. I leave on Thursday.

What's yours on Thursday

A working system and the means to keep it working — not a subscription.

What you keep

A running system on your own account. A back office with AI clerks in it. Each one works for one of your people, in a role, inside limits the system enforces, and that person reviews the work.
A trust budget for every clerk. Each starts on probation, with every run reviewed. Review drops to one run in two, then one in four, then one in ten, only as its acceptance rate proves out at 95% confidence. A rejection, failed check or refusal puts it back to reviewing everything, and so does a change of model. The efficiency grows only as fast as the quality is proven.
The map of that work. Every step, where it goes and why, and the check at each handoff — and which steps could move next, and what evidence would move them.
One team's work, set up for real. Their roles, their limits, their review routine, and what they know written down where their clerks use it.
A record of who taught it what. Every piece of guidance carries its author's name, and the system counts how often work that used it was approved.
The source code. Everything in NovaSuperSyn Office at the time: the Office app, the system behind it, Agent Desk and the Agent Registry.
A license to use and change it, for good. Inside your business, with no fee to keep using it.
The handbook. How to deploy, maintain, extend and troubleshoot it, written so your IT department can work from it.
People who can run it. Your IT person will have deployed a change alone. Your team will have taught their clerks something themselves.

What this isn't

Not a support contract. There's no help desk and no on-call. If you want more of my time later, you book a morning, or a standing one.
Not a subscription. You get the code as it stands that week. Later versions aren't included.
Not a rebuild of your existing systems. In three mornings we set up one team properly. Wiring it into the software you already run is the next piece of work, and your people will know how to start it.
Not AI left to decide alone. Every clerk starts with a person checking its work and earns less checking only on its record. What it may never do alone, you decide. This isn't blind automation; it's a workforce that earns its trust.
Not yours to resell. The license is for your business to use. It doesn't cover selling it, hosting it for others or installing it for them.

See it before we talk. Agent Desk is one part of what you'd be running. It takes a minute and needs no sign-up: give an AI clerk a job, set its limits, and watch the system stop it when it reaches too far.

Open Agent Desk

Before we start

Three mornings go quickly. These need to be in place before the Tuesday, so the first hour isn't spent waiting for a login.

Accounts, in your company's name

A Cloudflare account on the paid Workers plan. This is where it runs.
A GitHub organization. This is where your copy of the code and your records live.
A key for the AI models. It runs on open models, so you're never tied to one supplier.
A web address for it — a subdomain of your own is fine.

I'll send exact instructions for each when the week is booked. Running costs are yours and are small.

People, for all three mornings

Someone from IT who will look after it afterwards, with the access to create and change things in the accounts above.
Someone who does the work — who knows how the job is really done, including the parts nobody wrote down.
Someone who can decide what an AI clerk may do without asking, and what always needs a person.
One piece of work chosen to start with: enquiries, tickets, invoices, bills — something that happens every day.

In a small business these may be one or two people. That's fine.

Afterwards

Your team runs it without me. What keeps changing is everything around it: AI moves month to month, and every few months it can take on another step of your work. Someone has to watch the field for you, read what your own records are saying, and decide which step moves next. If you'd like that to be me, I keep a few standing mornings.

Your AI consultant, every week

A standing morning

$14,670

a month · one morning a week

The same weekday every week, four hours, done at noon. Month to month: either of us can stop with a month's notice. It ends when your map stops moving fast enough to need me.

I keep only a few, for companies that have done the three mornings.

Someone watching AI for you. Between mornings I keep watch with your business's interests in mind: new models and tools, new risks like scams aimed at your staff, and what your competitors can suddenly do. Each week you hear what matters to you, in plain words, before it catches you off guard.
Your company's own intelligence, at two levels. We read what your records are saying. For each person: how their AI clerks are doing and what would help. What a person improves stays theirs unless they choose to share it, and then it carries their name. For the whole company: bottlenecks, risks, and the patterns no one person can see. It comes only from your business, on your own account. Nothing is pooled with any other company's. Dashboards, reports and forecasting are work I've done for years; with AI, the question is what your data can now tell you that it couldn't before, and I help you find it.
What moved, and what slipped. Which steps earned a move to an AI clerk, which ones need to come back to a person.
The newest models, on your own work. When a better model comes out, we test it against work your team has already done, before anyone relies on it.
The next piece of work, mapped. One more of your team's jobs, step by step, with the check at each handoff.
Your team doing more of it each month. The goal is that they make these calls themselves. This isn't support: no help desk, no on-call, and I don't run your team.

$4,890

Another morning

Four more hours on whatever you choose: a second team, a connection to another system, a new role. Booked one at a time.